Our work has taken us on a journey through the world of sentiment analysis, but was limited to polarity classification. There are however several possible directions for future work emerging from the implementation of this thesis.
This thesis is based on word polarity (positive and negative). This is converted to either +1 or -1, which is discrete polarity. From this point, there is
another question that has not been investigated and answered in this thesis: ‘does it make any difference whether we use discrete and continuous polarity?’ Continuous polarity refers to a range of polarity that could be a real number (e.g. - 2, -1, 0, 1, and 2) or a floating point number (e.g. 0.10, 0.15, and 0.20).
To answer this question, an unsupervised learning algorithm could be used, where the learning algorithm does not require labelled datasets as the input data. The most classic unsupervised learning method is clustering. Clustering is a set of algorithms that analyses groups of data based only on information found in the data that describes the objects and their relationships. The goal of clustering is to determine the intrinsic grouping in a set of data (Tan et al., 2014). Clustering is needed for setting a threshold. A threshold is a parameter in which the upper and lower limits for the machine learning to interpret the range of polarity as positive/negative/neutral. Since the range polarity is not predefined nor should be defined by a person, clustering could provide the grouping significance of each polarity.
For example, a given range of polarity of 0 to 1, in which 0 is negative and 1 is positive. As the probability of value could be either 0 or 1, it is reasonable to put 0 to 0.49 as negative and 0.5 to 1 as positive. Nevertheless, this threshold setting does not reflect the nature of the data. Clustering, however, can group the data into clusters of range polarities and draw a threshold around the group. This could bring a negative threshold to < 0.3 and positive > 0.3 or negative < 0.7 and positive > 0.7 according to the clustering of the given dataset. The idea of clustering has been used by Maas et al. (2011)64 to assign labels to datasets for use to classify movie reviews. The idea is that a review that has a score which is less than or equal to 4 out of 10 is negative. On the other hand, a positive review has a score which is greater than or equal to 7 out of 10 while the rest are not included in the dataset. After gathering this dataset with binary polarity, the rest of the process is conducted using a support vector machine to classify the final output.
However, to answer the question above, the datasets should have both discrete and continuous polarity, the reason being that their final prediction could be used to compare whether or not their accuracy performance is the same. The
label of continuous polarity could be assigned using clustering or similar idea as Maas et al. (2011). An example of an ideal dataset is shown in Table 6.1. It is a sample of user reviews from TripAdvisor about Newcastle Airport Tourist Information65. Reviews Rate66from 1 to 5 (continuous polarity) Ideally of discrete polarity67
Hasnt changed one bit so they dont read reviews or care. PUB ALWAYS STINKS OF VINEGAR... if you dont wash and clean a bar properly ROTTEN BEER WILL STINK OF VINEGAR. Beer is warm and undrinkable
1 -1 (negative)
This airport needs to move into the 21st century, all very well having planes going all over the world but when you get back it is a bit ridiculous to wait nearly an hour for luggage. This ruins what was a great holiday being tired already after nearly 22 hours total travelling. Always convenient to get to and the flying out is very good but the coming back part is the letdown.
3 -1 (negative)
The first time I've flown from Newcastle for a few years. Everything went as planned the new bars and eateries upstairs were fine if not a little rowdy (stag and hen parties) that cant be helped.
4 1 (positive)
I had a great experience at the cabin, the staff were great and couldnt have been more helpful :) i would definalty choose the cabin again.
5 1 (positive)
Table 6-1: Example of dataset that has both discrete and continuous polarity
However, another process that could be used to answer this question is ‘meta-analysis’. Meta-analysis is a process that compares and combines quantitative results from several studies in the same area using statistics. By using meta-analysis, as many works as possible that are related to discrete and continuous polarity should be collected. The hypothesis for discrete and continuous polarity should be set as; there is no difference between using either discrete or continuous polarity. After that, the statistical method will be used to
65 http://www.tripadvisor.co.uk/Attraction_Review-g186394-d213735-Reviews-
Newcastle_Airport_Tourist_Information-Newcastle_upon_Tyne_Tyne_and_Wear_England.html
66 Based on the user’s rate in the website 67 This ideally is assigned manually by human
prove this hypotheses and their significance. A guide for choosing the appropriate statistics is presented in Figure 5.13.
Besides discrete and continuous polarity, there is another question of interest: will the accuracy improve when using a combination of sentiment classification and subjective classification? It has been observed in several studies that subjectivity classification may help to improve the performance of sentiment analysis. However, experiments conducted by Esuli and Sebastiani (2006a) and Zagibalov (2010) concluded that sentiment classification and subjective classification are separate tasks that simultaneously have to deal with a mixture of objective and subjective documents. This suggestion is led from sub-topics within sentiment analysis; they are sentiment classification and subjective classification. Sentiment classification is the task that classifies opinionated contexts as expressing a positive or a negative. On the other hand, subjective classification is a task that classifies a context as subjective or objective. Subjective refers to the opinion that expressions describe people’s sentiments or feelings toward entities (Liu, 2010). Objective concerns entities, events and their properties (Liu, 2010). This may be relevant to our work as our sentiment analysis focuses on both positive and negative contexts. Neutral sentiment tends to be much harder to identify as it requires the determination of the contexts of the message; for example, some content may have both subjective and objective senses. Handling these contents will therefore require the introduction of another classifier to identify the subjective and objective contexts.
In addition, there can be mixed sentiment contents. Many studies did not include mixed sentiment contents in the task due to the complexity of the ambiguously defined and typically inconsistent labelling (Bermingham, 2011). However, this does not mean that the mixed sentiment contents do not exist in the real-world. This task still remains for future work to identify how the mixed sentiment contents can be better identified using machine learning algorithms. Mixed sentiment content refers to the contents that have both positive and negative sentiments.
Nevertheless, content alone is inadequate for sentiment analysis. Humans use sociocultural data to interpret meanings from a piece of information. The most
obvious examples are sarcasm and persuasion. In order to understand sarcasm and persuasion in a microblog post, people use a combination of knowledge, experience and the history of interactions between different parties as the context.
However, to locate documents on a continuum, stretching from the extremely negative to the extremely positive is still a problem. Experiments in extreme polarity areas would require a special corpus that can be used to test the accuracy of the contents of a sentiment analysis. The corpus must follow the dimensional paradigm. It must use a specialised annotation scheme, which also needs a significant research effort with future work.
Another suggestion for further research is the real-time sentiment application for analysing some social networks such as Twitter and Facebook. The question arises: ‘Is it possible for using real-time sentiment application to detect review from the customers?’ This application will be useful for companies that are interested in how their customers perceive their products or services. Moreover, a language-independent approach would make it possible to monitor different national markets, while the absence of domain-dependency would allow a system to follow the twists of language use that occurs in real-life human communication. For example, the emerging of new topics of conversation with different styles of phrasing, speech and language are those which are difficult to predict.
There is another question that not answered in this thesis, which is ‘whether the number of rules effects the improvement of performance accuracy in Arbiter Tree (Chan and Stolfo, 1993)?’ To answer this question, each rule and each pairs should be tested independently. After that, their results could be compared for finding their effective and using for further study.
References
ABDEL-DAYEM, A. R. 2010. Proceedings of the 7th international conference on Image Analysis and Recognition - Volume Part II, Portugal. 2177026: Springer-Verlag, pp: 120-130.
ABDUL-MAGEED, M. and DIAB, M. 2014. Proceedings of the Language Resources and Evaluation Conference (LREC), Reykjavik, Iceland. pp: 1162-1169.
AGARWAL, A., BIADSY, F. and MCKEOWN, K. R. 2009. Proceedings of the 12th Conference of the European Chapter of the Association for Computational Linguistics, Athens, Greece. 1609069: Association for Computational Linguistics, pp: 24-32.
AHMAD, K. and ALMAS, Y. 2005. the 9th International Conference on Information Visualisation. pp: 363-368.
AISOPOS, F., PAPADAKIS, G. and VARVARIGOU, T. 2011. Proceedings of the 3rd ACM the Special Interest Group on Multimedia (SIGMM) international workshop on Social media, Scottsdale, Arizona, USA. 2072614: ACM, pp: 9-14. ALM, C. O., ROTH, D. and SPROAT, R. 2005. Proceedings of the conference on
Human Language Technology and Empirical Methods in Natural Language Processing. Association for Computational Linguistics, pp: 579-586.
ALTMAN, N. S. 1992. An introduction to kernel and nearest-neighbor nonparametric
regression. The American Statistician, Vol. 46, pp. 175-185.
AMAN, S. and SZPAKOWICZ, S. 2007. Text, Speech and Dialogue. Springer, pp: 196- 205.
AMBATI, V. 2008. Adv. MT Seminar Course Report. pp.
AMIRI, H. and CHUA, T.-S. 2012. Sentiment Classification Using the Meaning of
Words. In: JANNACH, D., ANAND, S. S., MOBASHER, B. and KOBSA, A.
(eds.) Intelligent Techniques for Web Personalization and Recommender
Systems: AAAI Technical Report WS-12-09. Palo Alto, California: The AAAI Press, Isbn: 9781577355748.
AUE, A. and GAMON, M. 2005. Proceedings of recent advances in natural language processing (RANLP). pp: 2-1.
BACCIANELLA, S., ESULI, A. and SEBASTIANI, F. 2010a. SentiWordNet. Available:
http://sentiwordnet.isti.cnr.it/
BACCIANELLA, S., ESULI, A. and SEBASTIANI, F. 2010b. Proceedings of the Seventh International Conference on Language Resources and Evaluation (LREC'10), Valletta, Malta. European Language Resources Association (ELRA), pp.
BALAHUR, A. 2013. 4th Workshop on Computational Approaches to Subjectivity, Sentiment and Social Media Analysis, Atlanta, Georgia. Association for Computational Linguistics, pp: 120-128.
BALDWIN, R. A. 2009. Use of maximum entropy modeling in wildlife research. Entropy, Vol. 11, pp. 854-866.
BALLAN, L., BERTINI, M., BIMBO, A., SEIDENARI, L. and SERRA, G. 2011. Event detection and recognition for semantic annotation of video. Multimedia Tools and Applications, Vol. 51, pp. 279-302.
BARAN, E. and WARRY, F. 2008. Statistical tests for comparing samples. Simple data analysis for biologists. WorldFish, Isbn: 9789995071011.
BARTLETT, M. S., LITTLEWORT, G., FRANK, M., LAINSCSEK, C., FASEL, I. and MOVELLAN, J. 2005. IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR). pp: 568-573.
BATTOCCHI, A., PIANESI, F. and GOREN-BAR, D. 2005. Dafex: Database of facial
expressions. Intelligent Technologies for Interactive Entertainment. Springer,
Isbn: 3540305092.
BEAR. 2015. Great Debut Film Performance By Eminem! [Online].
http://www.amazon.co.uk/. Available: http://www.amazon.co.uk/product- reviews/B00006FMGR/ref=cm_cr_pr_hist_5?ie=UTF8&filterBy=addFiveStar&s howViewpoints=0&sortBy=bySubmissionDateDescending.
BEHNEL, S., BRADSHAW, R., CITRO, C., DALCIN, L., SELJEBOTN, D. S. and
SMITH, K. 2011. Cython: The best of both worlds. Computing in Science &
Engineering, Vol. 13, pp. 31-39.
BEHNEL, S., BRADSHAW, R. and SELJEBOTN, D. 2008. Cython: C-extensions for Python.
BERMEJO, P., GÁMEZ, J. A. and PUERTA, J. M. 2011. Improving the performance of Naive Bayes multinomial in e-mail foldering by introducing distribution-based
balance of datasets. Expert Systems with Applications, Vol. 38, pp. 2072-2080.
BERMINGHAM, A. 2011. Sentiment analysis and real-time microblog search. Dublin
City University.
BERMINGHAM, A. and SMEATON, A. F. 2011. Proceedings of the Workshop on Sentiment Analysis where AI meets Psychology (SAAIP 2011). Asian Federation of Natural Language Processing, pp: 2-10.
BIFET, A. and FRANK, E. 2010. Proceedings of the 13th international conference on Discovery science, Canberra, Australia. Springer-Verlag, pp: 1-15.
BIKEL, D. M., SCHWARTZ, R. and WEISCHEDEL, R. M. 1999. An algorithm that
learns what's in a name. Machine learning, Vol. 34, pp. 211-231.
BIRD, S. 2006a. Proceedings of the COLING/ACL on Interactive presentation sessions. Association for Computational Linguistics, pp: 69-72.
BIRD, S. 2006b. Proceedings of the COLING/ACL on Interactive presentation sessions. Association for Computational Linguistics, pp: 69-72.
BIRD, S., KLEIN, E. and LOPER, E. 2009a. Accessing Text Corpora and Lexical
Resources. Natural Language Processing with Python. O'Reilly Media, Isbn:
9780596516499
BIRD, S., KLEIN, E. and LOPER, E. 2009b. Natural language processing with Python, O'Reilly, Isbn: 9780596516499.
BISHOP, C. M. 2006. Pattern recognition and machine learning, springer New York,
Isbn: 9780387310732.
BLACK, K. 2011. Sampling and Sampling Distribution. Business Statistics: For
Contemporary Decision Making. Wiley, Isbn: 9780470931462.
BLEI, D. M., NG, A. Y. and JORDAN, M. I. 2003. Latent dirichlet allocation. the
Journal of machine Learning research, Vol. 3, pp. 993-1022. BLITZER, J., DREDZE, M. and PEREIRA, F. 2007. ACL. pp: 440-447.
BRADLEY, M. M. and LANG, P. J. 1999. Affective norms for English words (ANEW): Instruction manual and affective ratings. Technical Report C-1, The Center for Research in Psychophysiology, University of Florida.
BREIMAN, L. 1996. Bagging predictors. Machine learning, Vol. 24, pp. 123-140.
BREIMAN, L. 2001. Random forests. Machine learning, Vol. 45, pp. 5-32.
BRILL, E. 1994. Proceedings of the twelfth national conference on Artificial intelligence, Seattle, Washington, USA. 199378: American Association for Artificial Intelligence, pp: 722-727.
BUTLER, E. 2010. Senior Seminar Conference, University of Minnesota, Morris. pp: 11-16.
CARIDAKIS, G., WAGNER, J., RAOUZAIOU, A., CURTO, Z., ANDRÉ, E. and KARPOUZIS, K. 2010. Multimodal Corpora: Advances in Capturing, Coding and Analyzing Multimodality. pp: 80.
CELIKYILMAZ, A., HAKKANI-TUR, D. and FENG, J. 2010. Spoken Language Technology Workshop (SLT). IEEE, pp: 79-84.
CHAKRABORTY, G., PAGOLU, M. and GARLA, S. 2014. Clustering and Topic
Extraction. Text Mining and Analysis: Practical Methods, Examples, and Case
Studies Using SAS. SAS Institute, Isbn: 9781612907871.
CHALOTHORN, T. and ELLMAN, J. 2014. Proceedings of the 8th International Workshop on Semantic Evaluation (SemEval 2014), Dublin, Ireland. Association for Computational Linguistics and Dublin City University, pp.
CHAN, P. K. 1996. An extensible meta-learning approach for scalable and accurate
inductive learning. Columbia University
CHAN, P. K. and STOLFO, S. J. 1993. The International Association for the Advancement of Artificial Intelligence (AAAI) workshop in Knowledge Discovery in Databases. pp: 227-240.
CHAN, P. K. and STOLFO, S. J. 1995. Conference on Knowledge Discovery and Data Mining (KDD). pp: 39-44.
CHAN, P. K. and STOLFO, S. J. 1997. On the accuracy of meta-learning for scalable
data mining. Journal of Intelligent Information Systems, Vol. 8, pp. 5-28.
CHANG, C.-C. and LIN, C.-J. 2011. LIBSVM: a library for support vector machines. ACM Transactions on Intelligent Systems and Technology (TIST), Vol. 2, pp. 27. CHATFIELD, C. 1983a. The design and analysis of experimetns - 1 Comparative
experiments. Statistics for Technology: A Course in Applied Statistics, Third
Edition. Taylor & Francis, Isbn: 9780412253409.
CHATFIELD, C. 1983b. The design and analysis of experimetns - 2 Factorial
experiments. Statistics for Technology: A Course in Applied Statistics, Third
Edition. Taylor & Francis, Isbn: 9780412253409.
CHEN, J. Y. and LONARDI, S. 2009. Biological Data Mining, Taylor & Francis, Isbn: 9781420086843.
CHEN, T. and KAN, M.-Y. 2013. Creating a live, public short message service corpus:
the NUS SMS corpus. Language Resources and Evaluation, Vol. 47, pp. 299-
335.
CHOW, S.-C. and LIU, J.-P. 2004. Designs for Clinical Trials. Design and Analysis of
Clinical Trials: Concepts and Methodologies. Wiley, Isbn: 9780471249856. CHOWDHURY, G. G. 2010. Natural language processing and information retrieval.
Introduction to Modern Information Retrieval. Neal-Schuman Publishers, Isbn: 9781856046947.
CHOY, M., CHEONG, M. L., LAIK, M. N. and SHUNG, K. P. 2011. A sentiment analysis of Singapore Presidential Election 2011 using Twitter data with census correction. arXiv preprint arXiv:1108.5520, Vol.
CIPRA, B. A. 1987. An introduction to the Ising model. American Mathematical
Monthly, Vol. 94, pp. 937-959.
COHEN, J. 1968. Weighted kappa: Nominal scale agreement provision for scaled
disagreement or partial credit. Psychological bulletin, Vol. 70, pp. 213.
CORRIGAN, J. 2008. The Oxford Handbook of Religion and Emotion, Oxford University Press, USA, Isbn: 9780195170214.
CRAMMER, K. and SINGER, Y. 2003. Ultraconservative online algorithms for
multiclass problems. The Journal of Machine Learning Research, Vol. 3, pp.
951-991.
CRISTIANINI, N. and SHAWE-TAYLOR, J. 2000. An introduction to support vector
machines and other kernel-based learning methods, Cambridge university press, Isbn: 0521780195.
CUFOGLU, A., LOHI, M. and MADANI, K. 2008. International Conference on Computer Engineering & Systems (ICCES). pp: 210-215.
CUNNINGHAM, H. 2002. Proceedings of the 40th Annual Meeting on Association for Computational Linguistic. pp: 168-175.
CUNNINGHAM, H., MAYNARD, D. and BONTCHEVA, K. 2011. Text Processing
with Gate (Version 6), Gate, Isbn: 9780956599315.
DANET, B., RUEDENBERG-WRIGHT, L. and ROSENBAUM-TAMARI, Y. 1997.
“HMMM…WHERE'S THAT SMOKE COMING FROM?”. Journal of
Computer-Mediated Communication, Vol. 2.
DASGUPTA, S., KALAI, A. T. and MONTELEONI, C. 2009. Analysis of perceptron-
based active learning. The Journal of Machine Learning Research, Vol. 10, pp.
281-299.
DAY, D., MCHENRY, C., KOZIEROK, R. and RIEK, L. 2004. International Conference on Language Resources and Evaluation. pp.
DENECKE, K. 2008. IEEE 24th International Conference Data Engineering Workshop (ICDEW) pp: 507-512.
DERCZYNSKI, L. 2013. Natural Language Toolkit: SVM-based classifier. Available:
http://www.nltk.org/_modules/nltk/classify/svm.html
DERKS, D., FISCHER, A. H. and BOS, A. E. 2008. The role of emotion in computer-
mediated communication: A review. Computers in Human Behavior, Vol. 24, pp.
766-785.
DEVITT, A. and AHMAD, K. 2007. Proceedings of the 45th Annual Meeting of the Association of Computational Linguistics. Association for Computational Linguistics, pp: 984-991.
DI EUGENIO, B. and GLASS, M. 2004. The kappa statistic: A second look. Computational linguistics, Vol. 30, pp. 95-101.
DOMHOFF, G. W. 2003. The Scientific Study of Dreams: Neural Networks, Cognitive
Development, and Content Analysis, Amer Psychological Assn, Isbn: 9781557989352.
DONG, Z. and DONG, Q. 2006. Hownet And the Computation of Meaning, World
Scientific Publishing Co., Inc., Isbn: 9812564918.
DONNE, J. 2013. Delphi Complete Poetical Works of John Donne (Illustrated), Delphi Classics, Isbn: 9781908909763.
DONNE, J. and ALFORD, H. 1839. The Works of John Donne: With a Memoir of His Life, Parker, Isbn: -.
DOWNEY, A. B. 2014. Relationships Between Variables. Think Stats. O'Reilly Media,
Isbn: 9781491907375.
DU, S. 2008. On the Use of Natural Language Processing for Automated Conceptual
Data Modeling. University of Pittsburgh.
DUAN, W., CAO, Q., YU, Y. and LEVY, S. 2013. System Sciences (HICSS), 2013 46th Hawaii International Conference on. IEEE, pp: 3119-3128.
DWORK, C., FELDMAN, V., HARDT, M., PITASSI, T., REINGOLD, O. and ROTH, A. 2015. Advances in Neural Information Processing Systems. pp: 2341-2349. EKMAN, P. and FRIESEN, W. 1978. Facial Action Coding System: A Technique for
the Measurement of Facial Movement. Consulting Psychologists Press, Vol. -. ELANGOVAN, M., RAMACHANDRAN, K. I. and SUGUMARAN, V. 2010. Studies